Brain tumor detection using GPT-NeoX-20B large language model-driven scaling wide residual neuron attention stage-by-stage net

Tumors arise due to the irregular growth of brain tissues. These issues cause severe damage to the brain activities, often causing death. The patient’s lifetime is increased by offering appropriate treatment at the beginning condition. However, the detection methodologies become difficult owing to the locations and shapes of tumors. Prompt generation is used in clinical decisions, and generates better outputs. Hence, this paper proposes the Scaling Wide Residual Neuron Attention Stage-by-Stage Net (SWideR-NasNet) for detecting brain tumors. The image is denoised by the Medav and Gaussian filters. The Power law transformations are used for image enhancement. The tumor region is segmented using Binary Cross-Entropy (BCE) and Dice loss-based Attention DeepLabv3+. The Deep Neural Network (DNN) and Local Directional Number Pattern (LDNP) features are extracted, and tumors are detected by the SWideR-NasNet. The detected output label is used to generate the prompt. The GPT-NeoX-20B Large Language Model (LLM) provides the text form of detected output using the prompt. The performance of SWideR-NasNet is assessed using accuracy, sensitivity and specificity, achieving respective values of 91.98%, 91.84% and 92.52%.